Abstract: In order to reduce and prevent the mine water inrush accidents，and to identify the source of water inrush quickly and accurately，this paper proposes a method based on MIV (Mean Impact Value) recognition algorithm of hybrid PSO-SVM source,to more effectively eliminate the underground water index information overlap between selected index systems better,so as to further improve the recognition accuracy of water.First,the PSO-SVM network is trained by using the water samples containing all the characteristic variables,and then the samples are added and subtracted respectively to a certain proportion to form a new sample into the trained network,according to the identification results obtaining the MIV value of each influence factor.Next,according to the principle of prior selecting high weight variables,the low weight variables are eliminated successively by judging the root mean square error to establish the optimal index system,which is fed back to the PSO-SVM for training and prediction.The selected sample is from Xinzhuangzi mine used to conduct 50 experiments,which is compared with the traditional PSO-SVM and other models.The results show that MIV-PSO-SVM model can measure the characteristic variables’ impact on the predicted results of the weighting more scientific and objectively,to build a more reasonable index system.The average accuracy of the model is 94.667%,the root mean square error is about 0.196 3,and the mean absolute error is about 3.413%,which significantly improves the average prediction accuracy and obviously decreases the root mean square error and the mean absolute error percentage,compared with other models.